Guojie Song
Current Organizations / Institutes
The organizations/institutes from the scholar's latest article
Research Subjects
Guojie Song has a total of 63 co-authors. Here are the top co-authors by number of shared publications:
- Fan Min
School of Computer Science and Software Engineering, Southwest Petroleum University, Chengdu, China
16 shared publications - Shulin Pan
Southwest Petroleum University
15 shared publications - Yinghe Wu
School of Geoscience and Technology, Southwest Petroleum University, Chengdu, China
5 shared publications Zhiliang Wang
School of Sciences of Southwest Petroleum University
5 shared publications- Qiong Xu
School of Computer Science and Software Engineering, Southwest Petroleum University, Chengdu, China
4 shared publications - Xinmin Zhang
Southwest Petroleum University 1 , School of Sciences, Chengdu, China
4 shared publications
Latest Publications
- 25 June 2026
- Published by Taylor & Francis in Petroleum Science and Technology
- p. 1-22
A drilling overflow prediction algorithm based on a TCN-Transformer-LSTM fusion network
- 25 June 2026
- Published by Taylor & Francis in Petroleum Science and Technology
- p. 1-22
A drilling overflow prediction algorithm based on a TCN-Transformer-LSTM fusion network
- Open Access
- 21 April 2026
- Published by MDPI AG in Mathematics
- Vol. 14 (8) , 1392
Residual Physics-Informed Neural Networks for Seismic Tomography with Multi-Source Prior Constraints
Seismic traveltime tomography is essential for constructing subsurface velocity models that underpin high-resolution imaging and inversion. Traditional ray- and eikonal-based methods are sensitive to the starting model and lack a unified, physically consistent framework to integrate seismic data with high-confidence prior information. PINN-based approaches offer flexible, grid-free inversion but often suffer from training instability and limited use of prior constraints. We propose a Multi-Source Prior-Guided Residual Physics-Informed Neural Network (MSP-ResPINNs) to address these limitations through two key technical advancements. First, MSP-ResPINNs integrates a Residual Network (ResNet) architecture with Sinusoidal Representation (SIREN) activation to replace standard MLPs, ensuring the robust capture of high-frequency velocity gradients. Second, the framework implements a unified loss function that rigorously enforces multi-source constraints, including well-logs and geological horizons. Numerical experiments demonstrate that MSP-ResPINNs accurately reconstructs sharp velocity contrasts and complex geological features compared with conventional PINN-based approaches. Among the tested variants, the multiplicative factorization consistently provides the most stable and physically consistent results, outperforming the additive factorization.
- 21 April 2026
- Published by MDPI AG in Mathematics
- Vol. 14 (8) , 1392
Open AccessResidual Physics-Informed Neural Networks for Seismic Tomography with Multi-Source Prior Constraints
Seismic traveltime tomography is essential for constructing subsurface velocity models that underpin high-resolution imaging and inversion. Traditional ray- and eikonal-based methods are sensitive to the starting model and lack a unified, physically consistent framework to integrate seismic data with high-confidence prior information. PINN-based approaches offer flexible, grid-free inversion but often suffer from training instability and limited use of prior constraints. We propose a Multi-Source Prior-Guided Residual Physics-Informed Neural Network (MSP-ResPINNs) to address these limitations through two key technical advancements. First, MSP-ResPINNs integrates a Residual Network (ResNet) architecture with Sinusoidal Representation (SIREN) activation to replace standard MLPs, ensuring the robust capture of high-frequency velocity gradients. Second, the framework implements a unified loss function that rigorously enforces multi-source constraints, including well-logs and geological horizons. Numerical experiments demonstrate that MSP-ResPINNs accurately reconstructs sharp velocity contrasts and complex geological features compared with conventional PINN-based approaches. Among the tested variants, the multiplicative factorization consistently provides the most stable and physically consistent results, outperforming the additive factorization.
- 11 December 2025
- Published by Springer Nature in Computational Geosciences
- Vol. 29 (6) , 53
Towards complex seismic layers: A vision Transformer approach to full waveform inversion
- 11 December 2025
- Published by Springer Nature in Computational Geosciences
- Vol. 29 (6) , 53
Towards complex seismic layers: A vision Transformer approach to full waveform inversion
- Open Access
- 28 February 2025
- Published by MDPI AG in Fractal and Fractional
- Vol. 9 (3) , 153
A Fractional Hybrid Staggered-Grid Grünwald–Letnikov Method for Numerical Simulation of Viscoelastic Seismic Wave Propagation
The accurate and efficient simulation of seismic wave energy dissipation and phase dispersion during propagation in subsurface media due to inelastic attenuation is critical for the hydrocarbon-bearing distinction and improving the quality of seismic imaging in strongly attenuating geological media. The fractional viscoelastic equation, which quantifies frequency-independent anelastic effects, has recently become a focal point in seismic exploration. We have developed a novel hybrid staggered-grid Grünwald–Letnikov (HSGGL) finite difference method for solving the fractional viscoelastic equation in the time domain. The proposed method achieves accurate and computationally efficient solutions by using a staggered grid to discretize the first-order partial derivatives of the velocity–stress equations, combined with Grünwald–Letnikov finite difference discretization for the fractional-order terms. To improve the computational efficiency, we employ a preset accuracy to truncate the difference stencil, resulting in a compact fractional-order difference scheme. A stability analysis using the eigenvalue method reveals that the proposed method confers a relaxed stability condition, providing greater flexibility in the selection of sampling intervals. The numerical experiments indicate that the HSGGL method achieves a maximum relative error of no more than 0.17% compared to the reference solution (on a finely meshed domain) while being significantly faster than the conventional global FD method (GFD). In a 500 × 500 computational domain, the computation times for the proposed methods, which meet the specified accuracy levels used, are only approximately 4.67%, 4.47%, 4.44%, and 4.42% of that of the GFD method. This indicates that the novel HSGGL method has the potential as an effective forward modeling tool for understanding complex subsurface structures by employing a fractional viscoelastic equation.
- 28 February 2025
- Published by MDPI AG in Fractal and Fractional
- Vol. 9 (3) , 153
Open AccessA Fractional Hybrid Staggered-Grid Grünwald–Letnikov Method for Numerical Simulation of Viscoelastic Seismic Wave Propagation
The accurate and efficient simulation of seismic wave energy dissipation and phase dispersion during propagation in subsurface media due to inelastic attenuation is critical for the hydrocarbon-bearing distinction and improving the quality of seismic imaging in strongly attenuating geological media. The fractional viscoelastic equation, which quantifies frequency-independent anelastic effects, has recently become a focal point in seismic exploration. We have developed a novel hybrid staggered-grid Grünwald–Letnikov (HSGGL) finite difference method for solving the fractional viscoelastic equation in the time domain. The proposed method achieves accurate and computationally efficient solutions by using a staggered grid to discretize the first-order partial derivatives of the velocity–stress equations, combined with Grünwald–Letnikov finite difference discretization for the fractional-order terms. To improve the computational efficiency, we employ a preset accuracy to truncate the difference stencil, resulting in a compact fractional-order difference scheme. A stability analysis using the eigenvalue method reveals that the proposed method confers a relaxed stability condition, providing greater flexibility in the selection of sampling intervals. The numerical experiments indicate that the HSGGL method achieves a maximum relative error of no more than 0.17% compared to the reference solution (on a finely meshed domain) while being significantly faster than the conventional global FD method (GFD). In a 500 × 500 computational domain, the computation times for the proposed methods, which meet the specified accuracy levels used, are only approximately 4.67%, 4.47%, 4.44%, and 4.42% of that of the GFD method. This indicates that the novel HSGGL method has the potential as an effective forward modeling tool for understanding complex subsurface structures by employing a fractional viscoelastic equation.
- 7 January 2025
- Published by Society of Exploration Geophysicists in Geophysics
- Vol. 90 (4) , T99-T108
An asymptotic local finite-difference method for fractional Laplacian quasidifferential operator
We develop a novel asymptotic local finite-difference (ALFD) method for solving the fractional Laplacian wave equation in the time domain. This numerical method uses the trigonometric function-based generating function to derive the quasidifference operator, facilitating the discretization of the fractional Laplacian without requiring fast Fourier transforms during simulations. Theoretical analysis and numerical experiments demonstrate that the ALFD method is accurate and efficient. Stability analysis using the eigenvalue method reveals that our method enjoys a relaxed stability condition (the minimum Courant number of 0.659492), providing greater flexibility in the selection of sampling intervals. Numerical dispersion anisotropy is significantly mitigated for various very small Q values (Q < 10). The numerical solutions obtained via the ALFD method march closely with the analytical solutions derived from Green’s function. The computation speed of ALFD is significantly higher than the analytical method and the original global finite-difference (GFD) method. For a 2D model with 400 × 400 grid points, the computational time required by the ALFD method is approximately 9.96%, 5.40%, 4.94%, and 4.60% of that required by the analytical method, and only 2.54%, 1.37%, 1.25%, and 1.15% of that required by the GFD method, respectively. Numerical examples illustrate the effectiveness of the ALFD method in simulating heterogeneous attenuating media, highlighting its potential as a valuable tool for studying inverse problems in viscoelastic media.
- 7 January 2025
- Published by Society of Exploration Geophysicists in Geophysics
- Vol. 90 (4) , T99-T108
An asymptotic local finite-difference method for fractional Laplacian quasidifferential operator
We develop a novel asymptotic local finite-difference (ALFD) method for solving the fractional Laplacian wave equation in the time domain. This numerical method uses the trigonometric function-based generating function to derive the quasidifference operator, facilitating the discretization of the fractional Laplacian without requiring fast Fourier transforms during simulations. Theoretical analysis and numerical experiments demonstrate that the ALFD method is accurate and efficient. Stability analysis using the eigenvalue method reveals that our method enjoys a relaxed stability condition (the minimum Courant number of 0.659492), providing greater flexibility in the selection of sampling intervals. Numerical dispersion anisotropy is significantly mitigated for various very small Q values (Q < 10). The numerical solutions obtained via the ALFD method march closely with the analytical solutions derived from Green’s function. The computation speed of ALFD is significantly higher than the analytical method and the original global finite-difference (GFD) method. For a 2D model with 400 × 400 grid points, the computational time required by the ALFD method is approximately 9.96%, 5.40%, 4.94%, and 4.60% of that required by the analytical method, and only 2.54%, 1.37%, 1.25%, and 1.15% of that required by the GFD method, respectively. Numerical examples illustrate the effectiveness of the ALFD method in simulating heterogeneous attenuating media, highlighting its potential as a valuable tool for studying inverse problems in viscoelastic media.
- 13 November 2024
- Published by Institute of Electrical and Electronics Engineers (IEEE) in IEEE Transactions on Geoscience and Remote Sensing
- Vol. 62 (01962892) , 1-16
ABA-FWI: Augmented Boundary Attention for Full Waveform Inversion
Deep learning full waveform inversion (DL-FWI) is an end-to-end and time-efficient high-resolution imaging technique for subsurface media. Popular methods are often plagued by location drift and significant velocity misfits at the stratigraphic boundaries. In this study, we propose an augmented boundary attention algorithm (ABA-FWI) to focus on the key boundary information. Regarding network composition, the wavelet convolution (WTconv) layer and the spatial attention module (SAM) are incorporated into the encoder and the decoder, respectively. The WTconv layer captures low frequencies by obtaining a large receptive field without suffering from over-parameterization. SAM extracts distinctive information by utilizing the inter-spatial relationship of features for resolution enhancement. For loss function design, our reflection coefficient tuned boundary (RCTB) loss introduces the reflection coefficient to adjust the gradient map weight of low-contrast areas. It focuses on boundary regions characterized by speed transitions to minimize errors. Results on OpenFWI, the SEG simulation, and the Marmousi II slice datasets show that our method is superior to state-of-the-art data-driven methods, especially on boundary details. The source code is available at https://github.com/FanSmale/ABA-FWI.
- 13 November 2024
- Published by Institute of Electrical and Electronics Engineers (IEEE) in IEEE Transactions on Geoscience and Remote Sensing
- Vol. 62 (01962892) , 1-16
ABA-FWI: Augmented Boundary Attention for Full Waveform Inversion
Deep learning full waveform inversion (DL-FWI) is an end-to-end and time-efficient high-resolution imaging technique for subsurface media. Popular methods are often plagued by location drift and significant velocity misfits at the stratigraphic boundaries. In this study, we propose an augmented boundary attention algorithm (ABA-FWI) to focus on the key boundary information. Regarding network composition, the wavelet convolution (WTconv) layer and the spatial attention module (SAM) are incorporated into the encoder and the decoder, respectively. The WTconv layer captures low frequencies by obtaining a large receptive field without suffering from over-parameterization. SAM extracts distinctive information by utilizing the inter-spatial relationship of features for resolution enhancement. For loss function design, our reflection coefficient tuned boundary (RCTB) loss introduces the reflection coefficient to adjust the gradient map weight of low-contrast areas. It focuses on boundary regions characterized by speed transitions to minimize errors. Results on OpenFWI, the SEG simulation, and the Marmousi II slice datasets show that our method is superior to state-of-the-art data-driven methods, especially on boundary details. The source code is available at https://github.com/FanSmale/ABA-FWI.
- 9 August 2024
- Published by Wiley in Geophysical Prospecting
- Vol. 72 (9) , 3495-3510
Decoupled approximate qP‐ and qSV‐wave equations in attenuated transversely isotropic media
Accurate seismic models with anisotropy and attenuation characteristics are crucial to accurately imaging subsurface structures. However, the anisotropic viscoelastic equations are complex and require significant computational resources. In addition, the single‐mode waves have been sufficient for most practical exploration needs. However, separating the qP‐ and qSV‐waves in anisotropic viscoelastic wavefields is challenging. Thus, we propose a new method to approximate and efficiently separate the qP‐ and qSV‐waves in attenuated transversely isotropic media. First, we obtain the decoupled approximate phase velocities of qP‐ and qSV‐waves by a curve‐fitting method. Consequently, based on the average and maximum relative error analysis, our approximate qP‐ and qSV‐wave phase velocities are more accurate than the existing approximations. Additionally, our approximations have broader applicability, resulting in acceptable errors during their application. Second, based on the approximate qP‐ and qSV‐wave phase velocities, we derive the corresponding qP‐ and qSV‐wave equations for a complete decoupling of the qP‐ and qSV‐wave components in transversely isotropic media. Third, to combine the attenuation and anisotropy characteristics, we incorporate the Kelvin–Voigt attenuation model and obtain the decoupled qP‐ and qSV‐wave equations in attenuated transversely isotropic media. Then, we use an efficient and stable hybrid finite‐difference and pseudo‐spectral method to solve the new decoupled qP‐ and qSV‐wave equations. Finally, several numerical examples demonstrate the separability and high accuracy of the proposed qP‐ and qSV‐wave equations. We obtain a qP‐wave wavefield entirely devoid of SV‐wave artefacts. In addition, the decoupled approximate qP‐ and qSV‐wave equations are accurate and stable in heterogeneous media with different velocities and attenuation. The decoupled, approximated qP‐wave and qSV‐wave equations proposed in this paper can effectively separate the qP‐wave and qSV‐wave components, resulting in fully decoupled qP‐ and qSV‐wave wavefields in attenuated transversely isotropic media.
- 9 August 2024
- Published by Wiley in Geophysical Prospecting
- Vol. 72 (9) , 3495-3510
Decoupled approximate qP‐ and qSV‐wave equations in attenuated transversely isotropic media
Accurate seismic models with anisotropy and attenuation characteristics are crucial to accurately imaging subsurface structures. However, the anisotropic viscoelastic equations are complex and require significant computational resources. In addition, the single‐mode waves have been sufficient for most practical exploration needs. However, separating the qP‐ and qSV‐waves in anisotropic viscoelastic wavefields is challenging. Thus, we propose a new method to approximate and efficiently separate the qP‐ and qSV‐waves in attenuated transversely isotropic media. First, we obtain the decoupled approximate phase velocities of qP‐ and qSV‐waves by a curve‐fitting method. Consequently, based on the average and maximum relative error analysis, our approximate qP‐ and qSV‐wave phase velocities are more accurate than the existing approximations. Additionally, our approximations have broader applicability, resulting in acceptable errors during their application. Second, based on the approximate qP‐ and qSV‐wave phase velocities, we derive the corresponding qP‐ and qSV‐wave equations for a complete decoupling of the qP‐ and qSV‐wave components in transversely isotropic media. Third, to combine the attenuation and anisotropy characteristics, we incorporate the Kelvin–Voigt attenuation model and obtain the decoupled qP‐ and qSV‐wave equations in attenuated transversely isotropic media. Then, we use an efficient and stable hybrid finite‐difference and pseudo‐spectral method to solve the new decoupled qP‐ and qSV‐wave equations. Finally, several numerical examples demonstrate the separability and high accuracy of the proposed qP‐ and qSV‐wave equations. We obtain a qP‐wave wavefield entirely devoid of SV‐wave artefacts. In addition, the decoupled approximate qP‐ and qSV‐wave equations are accurate and stable in heterogeneous media with different velocities and attenuation. The decoupled, approximated qP‐wave and qSV‐wave equations proposed in this paper can effectively separate the qP‐wave and qSV‐wave components, resulting in fully decoupled qP‐ and qSV‐wave wavefields in attenuated transversely isotropic media.
- 7 June 2024
- Published by Springer Nature in Acta Geophysica
- Vol. 73 (1) , 311-320
The waveform comparison of three fractional viscous acoustic wave equations
- 7 June 2024
- Published by Springer Nature in Acta Geophysica
- Vol. 73 (1) , 311-320
The waveform comparison of three fractional viscous acoustic wave equations
- 6 June 2024
- Published by Elsevier in Computers & Geosciences
- Vol. 190
SeismicTransformer: An attention-based deep learning method for the simulation of seismic wavefields
- 6 June 2024
- Published by Elsevier in Computers & Geosciences
- Vol. 190
SeismicTransformer: An attention-based deep learning method for the simulation of seismic wavefields